Browse Prior Art Database

Task Dispatching Using a Shadow Directory

IP.com Disclosure Number: IPCOM000043925D
Original Publication Date: 1984-Oct-01
Included in the Prior Art Database: 2005-Feb-05
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Hoevel, LW: AUTHOR [+8]

Abstract

To reduce the finite cache penalty it is proposed to make the task dispatching decisions of the operating system sensitive to the current cache contents (L1 and L2) and to dispatch that task which is most likely to use the current cache contents most efficiently. Given a set of putative starting addresses derived from TCB OLDPSW (Task Control Block Old Program Status Word), it is possible to determine which of these addresses has the longest sequence of cache resident lines by chaining successively through the first and second level entries of the shadow directory and by testing second level entries against the L1 directory to determine if they are resident therein. First level entries can be assumed to be resident in the L2 cache.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 64% of the total text.

Page 1 of 1

Task Dispatching Using a Shadow Directory

To reduce the finite cache penalty it is proposed to make the task dispatching decisions of the operating system sensitive to the current cache contents (L1 and L2) and to dispatch that task which is most likely to use the current cache contents most efficiently. Given a set of putative starting addresses derived from TCB OLDPSW (Task Control Block Old Program Status Word), it is possible to determine which of these addresses has the longest sequence of cache resident lines by chaining successively through the first and second level entries of the shadow directory and by testing second level entries against the L1 directory to determine if they are resident therein. First level entries can be assumed to be resident in the L2 cache. This technique can stop on the first second level entry which misses in the L1 cache or can be generalized to stop after a fixed number of accesses as long as one can continue to chain. The criteria for dispatching within the operating system can be based on any of the following: 1) Longest sequence of hits. 2) Most hits within a fixed length chain. 3) Search stopping condition specified via condition code (e.g., no first level entry), etc. 4) Longest first-level-chain in the shadow. A new instruction is required to implement this dispatching technique. The instruction accesses the shadow directory directly for the purpose of initiating the chain search and returning the relevant counts and con...